The authenticity of Hadith is vital for Muslims, as they are commanded to follow the example of Prophet Muhammad (PBUH).Hadith has two main components: the Matan, the textual content of a Hadith, and the Sanad, the chain of narrators who convey the Hadith. With the advancement of the Internet, accessing Hadith has become significantly easier, but this increases the risk of spreading false Hadith. This study explores the use of Natural Language Processing (NLP) for Hadith authenticity classification into three classes (Sahih, Hasan, and Da’if) and two classes (Accepted and Rejected), using both the Hadith full-text and Sanad-only text to compare them on the same task. It compares the performance of ArabicBERT against the traditional Machine Learning (ML) model Multinomial Naive Bayes (NB) and the deep learning (DL) model CNN-LSTM. The results show that ArabicBERT, fine-tuned using Optuna, performs the best in all of the configurations, In the 3-class configuration, ArabicBERT achieved an F1-score of 74.53\% for both the full-text and Sanad-only versions. In the 2-class configuration, it achieved 94.83\% (full-text) and 94.90\% (Sanad-only), showing better results compared to ARBERT in a previous study. The results show that using either full-text or Sanad-only text does not impact the performance of the models significantly. These results indicate that the use of ArabicBERT is effective in the task of Hadith classification, but the 3-class configuration can still be improved by handling the misclassification between Hasan and Da'if classes.